Distributed Database Systems - Database Systems - page 356

Rule 3: Continuous Operation

The system must be able to run continuously. There should be no need for a planned shut
down in order to carry out any function (for instance backup or tuning).

Rule 4: Location Independence (Transparency)

Users should not need to know where data is physically stored in order to access it; the
system should operate as if all the data resided at the local site. The distributed nature of
the database must be transparent to the end end users.

Rule 5: Fragmentation Independence

The system should support data fragmentation — it should be possible to partition a
given relation into fragments that are stored at different sites. Thus, data can be stored
where it is most frequently used. Network trafficking is therefore reduced. Fragmentation
should be transparent to the end users.
Example 1: Suppose that a large organization has employee records in an Employee
relation. The departments are at different localities. Records for each department are
stored at different sites in those respective departments. This can be easily facilitated in
Oracle or DB2 by partitioning the Employee table. A full discussion of table partitioning
is beyond the scope of this course. However, suffice it to say that that this is the technique
used by several leading DBMS suites to facilitate fragmentation independence.

Rule 6: Replication Independence

A given relation (or fragment of a relation) can be replicated at different sites. Replication
may improve access time and hence performance. The drawback however, is that at
update, all copies have to be updated simultaneously. Replication should be transparent
to the end users.

Rule 7: Distributed Query Processing

Distributed query processing must be facilitated among different sites. Records are
transmitted set (relation) at a time instead of record at a time.
Example 2: Suppose we have an international company, IBM, say, where employees
are stored in the relation Employee , fragmented in various countries, where there are
IBM offices. IBM Canada, issues the request: “Find all Jamaican male employees.” Then:

a. Suppose there are n records that satisfy this request. If the

system is relational, the query will involve tw o messages one
from Canada to Jamaica and one from Jamaica to Canada. If
the system is not relational, but record-at-a-time, the query
will involve 2n messages — n from Canada to Jamaica and n
from Jamaica to Canada.

b. Query optimization of the request occurs before execution

(record-at-a -time requests cannot be optimized).

Due to these two points, distributive database systems are usually relational.